Вы находитесь на странице: 1из 11

Weekly Progress Review

Manuel Ayala
12/21/2019
Major contributions of paper:
1. Introduces the T-S modeling and parallel distributed compensation (PDC) control approach
2. Shows how the approach can be applied to aircraft
3. Approach is modified to NASA F-15 nonlinear model and simulations are presented

T-S modeling and control (Mathematical formulation):


• TSFM is a specific mathematical form of representing non-linear dynamics. System dynamics are rewritten to fit:
𝑥ሶ 𝑡 = 𝐴 𝑥, 𝑢 𝑥 𝑡 + 𝐵 𝑥, 𝑢 𝑢(𝑡)
ቊ [1]
𝑦 𝑡 = 𝑥(𝑡)
• A and B are non-linear matrices
• C is taken to be identity
• D is zero
• All these assumptions are acceptable for this application because all of the aircraft states
are assumed to be directly measured

• The idea behind TSFM is to relabel the model’s nonlinearities of interest as so-called premise variables
• So, at any operating condition within a predefined range, the premise variables can be evaluated and Eq (1) is then
correct.
• Thus, in order to evaluate all premise variables, TSFM defines a set of fuzzy if-then rules corresponding to each premise
variables’ maximum and minimum over the predefined range of interest and every combination between premise
variables
• Each rule has corresponding linear system associated to it.
• The rules linear system are then weighted through membership functions and finally summed
• The result is a recreation of the nonlinear system, and although the TSFM is composed of a set of weighted summed
linear systems, the result is a nonlinear because the weightings are function of the states and inputs.
T-S modeling and control (Mathematical formulation):
• 𝑍𝑗 = 𝑃𝑟𝑒𝑚𝑖𝑠𝑒 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 (𝑇ℎ𝑒𝑠𝑒 𝑎𝑟𝑒 𝑡ℎ𝑒 𝑛𝑜𝑛 − 𝑙𝑖𝑛𝑒𝑎𝑟 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛𝑠 𝑤𝑒 𝑤𝑜𝑢𝑙𝑑 𝑙𝑖𝑘𝑒 𝑡𝑜 𝑐𝑎𝑝𝑡𝑢𝑟𝑒 𝑤𝑖𝑡ℎ𝑖𝑛 𝑇𝑆𝐹𝑀)
• 𝐷 = 𝐶𝑜𝑚𝑝𝑎𝑐𝑡 𝑟𝑒𝑔𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑙𝑎𝑛𝑡
• 𝑃 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑟𝑒𝑚𝑖𝑠𝑒 𝑣𝑎𝑟𝑖𝑏𝑎𝑙𝑒𝑠
• 𝑟 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑢𝑙𝑒𝑠
• 2𝑃 = 𝑇𝑜𝑡𝑎𝑙 𝑟𝑢𝑙𝑒𝑠
• 𝑊𝑖 = 𝑃𝑟𝑜𝑑𝑢𝑐𝑡 𝑜𝑓 𝑎𝑙𝑙 𝑚𝑒𝑚𝑏𝑒𝑟𝑠ℎ𝑖𝑝 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛𝑠 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑖𝑡ℎ 𝑟𝑢𝑙𝑒

• 𝐹𝑜𝑟 𝑟𝑢𝑙𝑒 𝑖, 𝑖𝑓 𝑍1 𝑡 = 𝑀𝑖1 … … 𝑍𝑝 𝑡 = 𝑀𝑖𝑝 , 𝑡ℎ𝑒𝑛:


𝑥ሶ 𝑡 = 𝐴𝑖 𝑥 𝑡 + 𝐵𝑖 𝑢(𝑡)
ቊ [2]
𝑦 𝑡 = 𝐶𝑖 𝑥(𝑡)

• The value of each premise variable (𝑍𝑗 ) can be written as a weighted combination of the extrema. The weights are determined through membership functions

𝑊𝑖 𝑍 𝑡 = ς𝑃𝑗=1 𝑀𝑖𝑗 𝑍 𝑡 [3]


• Normalizing 𝑊𝑖

𝑊𝑖 𝑍 𝑡
ℎ𝑖 𝑍 𝑡 = σ𝑟 [4]
𝑖=1 𝑊𝑖 (𝑍 𝑡 )
• Combining all the rules’ dynamics, the overall system becomes
𝑥ሶ 𝑡 = σ𝑟𝑖=1 ℎ𝑖 𝑍 𝑡 [ 𝐴𝑖 𝑥 𝑡 + 𝐵𝑖 𝑢(𝑡)]
ቐ [5]
𝑦 𝑡 = σ𝑟𝑖=1 ℎ𝑖 𝑍 𝑡 [ 𝐶𝑖 𝑥(𝑡)]

• Calculation of the premise variables and the weights through the membership functions exactly recreates the non-linear system it is representing over the compact region D.
Parallel distributed compensation modeling and control (Mathematical formulation):
• The idea behind PDC: for each of the model rules, a control rule is created with the same conditions, which implements a controller
• The “if” statements are the same, with the same premise variables as the model rules, but the “then” statements defines a controller So the output will not be a crisp value
but a command for another controller?
• For the purpose of stabilizing the aircraft, where all the states can be measured, a state feedback controller has been implemented for each rule
• Using state feedback, the control rule can be expressed as follows:

𝑖𝑓 𝑍1 𝑡 = 𝑀𝑖1 … … 𝑍𝑝 𝑡 = 𝑀𝑖𝑝 , 𝑡ℎ𝑒𝑛: 𝑢 𝑡 = −𝐹𝑖 𝑥(𝑡)


Don’t think it is correct.
• The overall control input with weightings becomes:

σ𝑟𝑖=1 𝑊𝑖 𝑍 𝑡 𝐹𝑖 𝑥 𝑡
𝑢 𝑡 =− σ𝑟𝑖=1 𝑊𝑖 𝑍 𝑡
= − σ𝑟𝑖=1 ℎ𝑖 𝑍 𝑡 𝐹𝑖 𝑥(𝑡) [6]

• Leading to a closed loop system


𝑥ሶ 𝑡 = σ𝑟𝑗=1 ℎ𝑗 𝑍 𝑡 σ𝑟𝑖=1 ℎ𝑖 𝑍 𝑡 [ 𝐴𝑖 − 𝐵𝑖 𝐹𝑗 ]𝑥(𝑡)
ቐ [7]
𝑦 𝑡 = σ𝑟𝑖=1 ℎ𝑖 𝑍 𝑡 [ 𝐶𝑖 𝑥(𝑡)]

• Controller gains are numerically calculated using linear matrix inequalities method (LMI).
PDC Failure Response
• 2 types of failure: Actuator and plant failure
• Actuator failure is done trough locking control surfaces in place and taking away any control influence they have into the system. Equivalent o B-Matrix failure
• Plant failures correspond to failures in the airframe which are represented by changing the pitching stiffness for the longitudinal tests and changing yaw stiffness as well as the rudder
effectiveness for the lateral tests.

Longitudinal Model 1st simulation :


• Pitching stiffness changed to maximum and minimum
• 3 angles of attack (-4, 5, 10)
• Failure occurs at 0.1 sec
• No actuators
Longitudinal Model 2nd simulation:
• Disturbance rejection of the control system
• Vertical gust introduced at 5 sec
• Gust is 0.2 sec sine impulse with a maximum of 40 ft/s amplitude
• No actuators
• Failure at 3 sec
Lateral Model 1st Simulation:
• Failure occurs at 1.5 sec
• Failure of left Aileron. Losing total control influence. Locking it a 0 deg
• Failure of one vertical tail (Rudder). Represented by reducing the rudder effectiveness
• Bank angle initial conditions [ 5 to 25 deg]
Lateral Model 2nd Simulation:
• Same simulations as before but includes a lateral gust
• Actuators failure at 4 sec
• Vertical gust introduced at 7 sec
• Gust is 0.2 sec sine impulse with a maximum of 50 ft/s amplitude
Future reading/ Literature Review

He introduces TSFM and PDC

Fault-tolerant adaptive fuzzy control of


a linear aircraft model

More details of the TSFM


mathematical formulation

Petr Husek, Kashyapa Narenathreyas. “Aircraft longitudinal Another application of TSFM and PDC for
motion control based on Takagi–Sugeno fuzzy model”. another aircraft. Also, include FC type 2
2016. Applied soft computing.
Future reading/ Literature Review

HAI-JUN RONG, et al. “Adaptive Fuzzy Fault-Tolerant


Controller for Aircraft Autolanding under Failures”. 2007. Different approach to fault-tolerant control
IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC using fuzzy logic system
SYSTEMS

Different approach to fault-tolerant control


using fuzzy logic system

Xiang Yu, Yu Fu, Xiaoyan Peng. “Fuzzy Logic Aided Fault- Different approach to fault-tolerant control
Tolerant Control Applied to Transport Aircraft Subject to using fuzzy logic system
Actuator Stuck Failures”. 2018. IEEE Transactions on fuzzy
systems

Вам также может понравиться